Prediction Of Antarctic Meteorological Data Based On ARIMA-LSTM-SAM Model

Time series exist in all aspects of the objective world,and many data can be regarded as time series data.With the rapid development of computer technology and statistical means,the methods of analyzing time series are changing with each passing day.Nowadays,more and more researchers pay attention to the prediction of time series.Prediction is the basis of decision-making,so time series prediction is always a very practical research field,which has important reference value for decision-makers.Decision-making is the continuation of prediction.The purpose of prediction is to provide a basis for decision-making.The more scientific and reasonable the prediction,the more correct and reliable the decision-making.Although many researchers have done a lot of research on time series prediction,there is still a need to improve the prediction performance and accuracy of the model,especially with the rapid development of machine learning algorithms.In view of the fact that the prediction effect of time series can not meet the precision requirement of practical application,this paper presents a new model combining traditional analysis model with machine learning model,so the prediction precision of time series can be improved.In order to further improve the prediction effect of time series models,this paper studies some models based on traditional mathematical statistics and some models based on machine learning algorithm,several models are used to model and predict the time series data,and the error analysis is compared,and a lot of analysis is made in the selection of time series data.Because based on the idea of divide and conquer of complex problems and the method of gradual optimization of machine learning,the originally complex sequence can be decomposed into many simpler,more regular and stable subsequences,then,the corresponding methods are used to analyze and predict these sub-sequences.Finally,the prediction results of several sub-sequences are fused to form the final prediction of the original sequence.In this paper,we use HP HIGH-PASS FILTER to decompose the series into linear trend term and nonlinear periodic term,then use Arima model to model and forecast the trend term,and use machine learning fusion model to model and forecast the periodic term,the prediction results of the two terms are integrated to get the final prediction results of the original sequence.Compared with the previous research,the innovation of this paper is: Firstly,we use the idea of divide-and-conquer of complex sequence to decompose the sequence,and add a layer of self-attention in the common Long Short-Term Memory model,a new model is proposed innovatively——ARIMA-LSTM-SAM model.Compared the forecasting effects of ARIMA-LSTM-SAM model and other models on air temperature series in detail.Secondly,the model proposed in this paper is applied to other types of data,the experimental results show that the proposed model prediction accuracy is high,indicating that the model has a good generalization performance.